Egypt Has Launched the KARNAK Arabic LLM, and Cairo Has Just Become North Africa's Most Credible AI Capital
Egypt's National Council for Artificial Intelligence (NCAI), together with Cairo University and the Information Technology Institute,...
Egypt Has Launched the KARNAK Arabic LLM, and Cairo Has Just Become North Africa's Most Credible AI Capital
Egypt's National Council for Artificial Intelligence (NCAI), together with Cairo University and the Information Technology Institute, has officially launched KARNAK, the country's first sovereign Arabic foundation model. The announcement, timed to the opening of Ai Everything MEA 2026 in Cairo on 22 April, positions Egypt as the most serious AI capital in North Africa, and it anchors the country's stated ambition under Digital Egypt 2030 to raise the ICT sector's share of GDP to 7.7 percent by the end of the decade.
KARNAK, named after the temple complex in Luxor, is a 180 billion parameter mixture-of-experts foundation model trained on a curated Arabic corpus with heavy weighting toward Egyptian and North African dialects. It is the first Arabic model to be trained primarily on Egyptian dialect data, and it is explicitly engineered for use across Egyptian public services, media, and education, rather than as a general Gulf Arabic product.
What KARNAK Actually Is
The model is a mixture-of-experts architecture with 180 billion total parameters and approximately 32 billion active per forward pass. It was trained across a hybrid cluster spanning the Egyptian Academy for Scientific Research and Technology and public sector compute procured under the Huawei and Silatech partnership announced in late 2025. Training used a 4.2 trillion token corpus, roughly 62 percent of which is Arabic, with a heavy emphasis on Egyptian dialect sources, North African Arabic, and Modern Standard Arabic.
Advertisement
The launch weights are open, under an Egyptian-defined licence that permits non-commercial academic use and commercial use subject to NCAI approval. That is a deliberate intermediate posture between fully open models like TII's Falcon and closed Gulf models like HUMAIN's internal foundation work. Egypt wants its talent base to build on KARNAK, while keeping strategic commercial uses under government oversight.
KARNAK is not just a technical project. It is a sovereignty project. Every Arabic speaker in North Africa deserves an AI model that understands how they actually talk, and Egypt has the talent base to build it.
By The Numbers
180 billion180 billion total parameters in the mixture-of-experts architecture, with approximately 32 billion active per forward pass.
4.2 trillion4.2 trillion tokens used in training, 62 percent of which are Arabic, per NCAI's technical report.
108 million108 million Arabic speakers addressable in Egypt alone, with another 110 million across North Africa and the Levant.
7171 percent ICT penetration in Tunisia, Egypt's closest Maghreb peer on AI capability, per the same diagnostic.
1616 African countries rolling out national AI strategies in 2026, including Egypt, Tunisia, and Morocco, according to Zawya.
Why Egypt Is Suddenly Serious
Egypt has had a long-running credibility gap on AI. The talent base was obvious: Cairo University, Ain Shams, and Alexandria University have produced top Arab AI researchers for a decade. The missing pieces were compute, capital, and a clear state strategy. All three have arrived in the past 12 months.
Compute arrived via the Huawei and Silatech infrastructure deal and a modest but capable public sector cluster at the Egyptian Academy for Scientific Research and Technology. Capital arrived through a combination of Gulf investment, EU development cooperation on responsible AI, and domestic funding under Digital Egypt 2030. State strategy arrived in the form of the Egypt National AI Guidelines, published earlier in 2026 and already cited by the MENA Observatory as a regional benchmark.
The KARNAK launch is the moment Egypt stops being a talent exporter and starts being a full AI producer. The policy, the compute, and the model are now aligned.
How KARNAK Fits the MENA Model Landscape
KARNAK does not try to compete head-on with the largest Gulf and global models. It is positioned as the default model for Egyptian and North African use cases, with a clear sovereignty claim and a dialect focus that Gulf-first models do not match.
Model
Origin
Size
Focus
Licensing
KARNAK
Egypt, NCAI
180B total, 32B active MoE
Egyptian and North African Arabic
Open weights, commercial with NCAI approval
Falcon 3
UAE, TII
Up to 180B dense
Gulf Arabic and multilingual
Open weights, permissive licence
Jais
UAE, G42
Up to 70B dense
Gulf Arabic foundation
Open weights
Fanar-2
Qatar, QCRI
85B dense
Qatari plus Gulf Arabic
Research and sector use
Noor (Ai71)
UAE, G42
Specialist summarisation
Arabic text summarisation
Managed service
HUMAIN foundation
Saudi Arabia
Multi-model marketplace
Saudi enterprise
Commercial, restricted
The differentiation is real. KARNAK is the only serious 180-billion-parameter Arabic model trained primarily on Egyptian and North African dialect, and that is a meaningful competitive moat across 220 million North African and Levantine Arabic speakers.
What It Unlocks for Egypt and the Maghreb
Three immediate applications are planned. Each is more consequential than a typical LLM launch because KARNAK is explicitly designed to plug into government services.
Public service Arabic chat: a national assistant for citizen interactions across ministries and governorates.
Education tooling: Egyptian dialect tutoring and curriculum generation for public schools, in partnership with the Ministry of Education.
Media and publishing: automated subtitling, translation, and content localisation for Egyptian cinema, television, and publishing.
Legal and judicial support: Egyptian legal research, case analysis, and drafting support for the Egyptian judiciary.
Healthcare communication: patient-facing education and triage prompts in Egyptian Arabic.
A fifth application, expected to follow later in 2026, is cross-Maghreb deployment for Morocco and Tunisia's public services, subject to local governance agreements. Morocco's Al Jazari Institutes are already positioning to work with KARNAK as part of the country's own 2030 strategy, and Tunisia's AI Institutes are likely to follow.
What Global Vendors and Regulators Should Notice
KARNAK also changes the negotiating position of North African governments with global AI vendors. When a country can point to a credible sovereign model, its leverage in procurement discussions with OpenAI, Anthropic, Google, and Microsoft improves materially. Egypt has just acquired that leverage, and Morocco and Tunisia will benefit from the KARNAK footprint in their own negotiations.
The AI in Arabia View: The KARNAK launch is the most consequential North African AI announcement of 2026, and it reshapes the MENA model landscape. Egypt now has a credible sovereign foundation model, an operational AI council, a published governance framework, and a clear path into public service deployment. That combination is unique in North Africa and it makes Cairo the most serious AI capital west of the Gulf. We expect KARNAK to be picked up by Morocco's Al Jazari Institutes and Tunisia's AI Institutes within 12 months, creating a Maghreb-plus-Egypt deployment footprint that operates meaningfully independently of Gulf or global models. For Gulf actors, KARNAK is a useful counterweight to hyperscaler dependency and a natural partner for Arabic-specific research. For global vendors, the negotiating table has moved. Egypt, Morocco, and Tunisia will bring sovereign models to the room when they talk about cloud, compute, and AI services, and that will change the economics of those deals. The sovereignty wave is real, and Cairo has joined it at the front.
AI Terms in This Article6 terms
LLM
A large language model, meaning software trained on massive text data to generate human-like text.
foundation model
A large AI model trained on broad data, then adapted for specific tasks.
tokens
Small chunks of text (words or word fragments) that AI models process.
parameters
The internal settings an AI model learns during training. More parameters generally means more capable.
benchmark
A standardized test used to compare AI model performance.
leverage
Use effectively.
Frequently Asked Questions
What is KARNAK and who built it?
KARNAK is Egypt's first sovereign Arabic foundation model, a 180 billion parameter mixture-of-experts system built by Egypt's National Council for Artificial Intelligence with Cairo University and the Information Technology Institute. It was launched on 22 April alongside Ai Everything MEA 2026 in Cairo.
How is KARNAK different from Falcon or Jais?
KARNAK is trained primarily on Egyptian and North African Arabic rather than Gulf Arabic, and its target use cases are Egyptian public services, education, and media. Falcon and Jais are more Gulf-oriented. The parameter scale is comparable, but the dialect focus and deployment posture are distinct.
Is KARNAK open source?
The weights are open under an Egyptian-defined licence that permits non-commercial academic use. Commercial use requires NCAI approval. That is an intermediate posture between fully open Gulf models and closed Saudi ones, designed to support Egyptian talent while keeping strategic commercial uses under government oversight.
Will KARNAK be used in Morocco and Tunisia?
Very likely. Morocco's Al Jazari Institutes and Tunisia's AI Institutes are expected to adopt KARNAK for public service and education deployments within 12 months, subject to local governance agreements. That would create a Maghreb-plus-Egypt AI deployment footprint operating independently of Gulf and global models.